Battery intelligence

Every site battery scored from real outage behaviour, inventory age and load — producing a traffic-light replacement queue for the NOC and field maintenance instead of calendar-based guesswork.

1,754

sites covered by battery scoring in the live CETIN SK deployment

BUT score

battery usability from real discharge behaviour during actual outages

8 weeks

replication pilot pattern for a new region or operator (plus 4 weeks scale-out)

What it does

For every site, the system continuously answers two questions the maintenance plan only guesses at: how long will this battery actually hold during an outage, and which batteries should be replaced next. It combines:

• Real discharge behaviour — measured battery performance during actual mains outages (the BUT score), not nameplate capacity

• Inventory and age — battery models and installation dates with age-based decay

• Outage history — how often and how long each site actually loses power

The result is a traffic-light replacement queue: red sites that will fail the next real outage, amber sites to plan, green sites that calendar-based replacement would have swapped for nothing.

At a glance

Status: Live

Deployed at: CETIN SK — production

Domain: Energy

Built with: ELTEK telemetry · Outage history · Inventory · BitSwan pipelines

Why it matters

Batteries are replaced either too late (a site goes dark during a grid event, customers and SLAs pay) or too early (capex is spent on cells that had years left). Scoring from observed behaviour replaces both errors with a ranked list — and during a live power outage, the same scores tell the NOC which sites in the affected region will fall over first, turning generator dispatch into triage with a clock on it.

How it works

BitSwan pipelines correlate ELTEK power telemetry, the power-outage event stream, and inventory data per site; scoring models update as new outages provide fresh evidence. NOC L2 and field maintenance consume the same queue — one prioritises during events, the other plans replacements between them.

Why it replicates

The replication pattern takes the architecture and scoring logic as-is: an 8-week pilot in one region, then a 4-week scale-out. An earlier feasibility assessment had rated this use case low for lack of data — the CETIN SK deployment proved the data was there all along; it just had never been joined.